Title: Bayesian mixture item response modeling in the presence of noncompliers
Authors: Kensuke Okada - The University of Tokyo (Japan) [presenting]
Abstract: Detecting noncompliers who do not follow questionnaire instructions is one of the major challenges in survey research. We propose a Bayesian mixture item response theory modeling approach to model and detect these noncompliers based on their answers to the questionnaire items. For this purpose, previously proposed latent variable models for response styles and for reverse-coded items were extended as a mixture model for both compliers and noncompliers. In order to fit the Bayesian model to the data, the Hamiltonian Monte Carlo algorithm was used to efficiently draw random samples from the joint posterior distribution. The proposed method was applied to a secondary dataset of psychological surveys. Our survey contained several instructional manipulation check items that served as indicators of noncompliance. The results showed that the proposed modeling approach provides better predictive fit and better interpretability than ordinary models that do not consider the effects of noncompliance. Our findings highlight the importance of taking the existence of noncompliers into account, particularly in this age of online surveys.